This paper presents a new algorithm of the singular value decomposition (SVD) in the wavelet domain for ground penetrating radar (GPR) to remove direct coupled waves. In fact, direct coupled waves commonly disturb the reflecting waves from underground targets. Besides, the amplitude and energy of direct coupled waves are large, which reduces the resolution of the images to the targets and adversely affects the subsequent image interpretation work. The GPR signal is decomposed into several levels by Wavelet to obtain approximation components and detailed components of each level. The information of targets is contained in big eigenvalues of detail components, while the direct coupled waves are contained in small ones. Therefore, the SVD in the wavelet domain can reduce the misjudgment of effective signals and improve the signal to noise ratio (SNR) of GPR signals. The simulated and field GPR data show that the SVD in the wavelet domain denoising method has better results for direct coupled wave removal than the traditional methods, which validates the effectiveness of the proposed denoising method.